Development. biologically-inspired computing. lecture 16. Informatics luis rocha x x x. Syntactic Operations. biologically Inspired computing

Size: px

Start display at page:

Download "Development. biologically-inspired computing. lecture 16. Informatics luis rocha x x x. Syntactic Operations. biologically Inspired computing"

Lester Curtis
5 years ago
Views:

1 lecture 16 -inspired S S2 n p!!! 1 S Syntactic Operations al Code:N Development x x x 1 2 n p S

Sections I485/H400 course outlook Assignments: 35% Students will complete 4/5 assignments based on algorithms presented in class Lab meets in I1 (West) 109 on Lab Wednesdays Lab 0 : January 14 th

2 Sections I485/H400 course outlook Assignments: 35% Students will complete 4/5 assignments based on algorithms presented in class Lab meets in I1 (West) 109 on Lab Wednesdays Lab 0 : January 14 th (completed) Introduction to Python (No Assignment) Lab 1 : January 28 th Measuring Information (Assignment 1) Graded Lab 2 : February 11 th L-Systems (Assignment 2) Graded Lab 3: March 25 th Cellular Automata & Boolean Networks (Assignment 3) Due: April 1 st Lab 4: April 8 th Genetic Algorithms (Assignment 4) Due: April 22 nd

Readings until now Class Book Nunes de Castro, Leandro [2006].

Chapman & Hall. Chapters 1, 2, 3.1-3.4, 7.1-7.5, 8.1-8.2, 8.3.10 Lecture notes Chapter 1: What is Life?

Chapter 4: Self-Organization and Emergent Complex Behavior posted online @

3 Readings until now Class Book Nunes de Castro, Leandro [2006]. Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications. Chapman & Hall. Chapters 1, 2, , , , Lecture notes Chapter 1: What is Life? Chapter 2: The Logical Mechanisms of Life Chapter 3: Formalizing and Modeling the World Chapter 4: Self-Organization and Emergent Complex Behavior posted Other materials Flake s [1998], The Computational Beauty of Nature. MIT Press Chapters 20

final project schedule ALIFE 15 Projects

edu/alife14nyc/ 8 pages (LNCS proceedings

4 final project schedule ALIFE 15 Projects Due by May 4 th in Oncourse ALIFE 15 (14) Actual conference due date: pages (LNCS proceedings format) D= Preliminary ideas overdue! Individual or group With very definite tasks assigned per member of group

5 the genetic code at work DNA Replication/Synthesis P Reproduction < DNA Polymerase P Transcription < RNA Polymerase P Translation < Ribosome P Coupling of AA s to adaptors < Aminoacyl Synthetase Translation Transcription Figures from Eigen [1992]. Steps Towards Life.

6 Complexity galore genetic information at work Many intermediate levels subject to control Development, environment, epigenetic, operational

Welmer et al 2006, PLOSGEN) Expression of genes controlled by the products of other genes

7 Informatics regulatory networks genetic information at work Gene regulatory network controlling early Arabidopsis flower development (F. Welmer et al 2006, PLOSGEN) Expression of genes controlled by the products of other genes (proteins and RNAs) Regulatory networks whose (sel-organizing) dynamics are modulated by various factors, inlcuidng information stored in DNA tape

information transmission Offspring with a new genotype Mitosis Eukaryotic cell separates its duplicated genotype into two identical halves

8 Genetic information transmission Meiosis Diploid cell's genome is replicated once and split twice produces four haploid (germ) cells each with half the chromosomes Sexual reproduction combines germ cells from two individuals to produce diploid (zygocyte) cells Vertical genetic information transmission Offspring with a new genotype Mitosis Eukaryotic cell separates its duplicated genotype into two identical halves somatic cells in multicellular organisms Horizontal genetic information expression development Crossovers may occur in meiosis Cell division

9 meiosis replication Genotype DNA Inherited variation transcription genetic information at work translation (code) genotype/phenotype RNA amino acid chains Germ cell line Can this be generalized any further? Development, regulation environmental ramifications mitosis phenotype organism

modeling genetic-based (open-ended) evolution History of evolutionary computation Evolutionary Operation Box (1957) Perturbations to continuous variables followed by selection to improve industrial

10 modeling genetic-based (open-ended) evolution History of evolutionary computation Evolutionary Operation Box (1957) Perturbations to continuous variables followed by selection to improve industrial productivity Evolution Strategies Rechenberg (1960 s), Schweffel (1970 s) To optimize real-valued parameters in wind-tunnel experiments Real-valued genotypes under variation and selection Evolutionary Programming Fogel, Owens, and Walsh (1966) Evolution of tables of state-transition functions (diagrams) under mutation and selection Artificial ecosystems Conrad and Pattee (1970) Population of artificial cells evolving with genotype and phenotype Other early evolution-inspired algorithms and models Barricelli CA-like model(1957), game-strategy model (1963) Symbiogenetic evolution Friedman (1957, 1959), Bledsoe (1961), Bremmermann (1962) Genetic Algorithms John Holland (1960 s and 1970 s) Adaptation in Natural and Artificial Systems, University of Michigan Press, (MIT Press, second edition 1992)

via genetic algorithms optimization Search algorithms based on the mechanics of Natural Selection Holland, Conrad, Fogel Based on distinction between a machine and a description of a machine Solution

11 via genetic algorithms optimization Search algorithms based on the mechanics of Natural Selection Holland, Conrad, Fogel Based on distinction between a machine and a description of a machine Solution alternatives for optimization problems f Objective function df dx Random Search directionless 0 df dx 0 x Enumerative Search Search point by point Direct analysis depends on Knowing the function Existence of derivatives continuity hill-climbing hop on the function and move along the steepest direction until a local extrema is found

12 examples fitness landscapes

probabilistic transition rules Statistical bias Not deterministic rules Advantages Samples the space widely like an enumerative or

13 genetic algorithms Work with an encoding of the parameter set, not the parameters themselves solution alternative encoded as descriptions Search a population of points in parallel Not a single point Uses objective function directly Not derivatives Uses probabilistic transition rules Statistical bias Not deterministic rules Advantages Samples the space widely like an enumerative or random algorithm, but more efficiently Directed search Not so easily stuck on local extrema Population search Variation mechanisms to search new points

of Natural Selection Based on distinction

14 artificial genotype/phenotype mapping Search algorithms based on the mechanics of Natural Selection Based on distinction between a machine and a description of a machine Solution alternatives for optimization problems Genotype DNA Inherited variation computational evolution transcription translation (code) development RNA amino acid chains Traditional Genetic Algorithm Genotype environmental ramifications phenotype organism S S n 2 p!!! 1 S Variation Code: φ code x x x 1 2 n p Phenotype Selection

{0, 1} GA s exploit coding similarities Searches the code space, not the

15 In genetic algorithms Solution space encoded as finite-length string over a finite alphabet E.g. {0, 1} GA s exploit coding similarities Searches the code space, not the solution space coding code Traditional Genetic Algorithm Genotype S S 2 n p!!! 1 S Code: φ x1 x 2 x n p Phenotype Variation Selection

16 The workings genetic algorithms 1) Generate Random population of bit-strings Chromosomes/Genotypes A candidate solution to a problem Genes Single or short blocks of adjacent bits Allele Actual value of a gene 2) Evaluate Fitness Function for each decoded solution 3) Reproduce next generation Genotypes of solutions with higher fitness value reproduce with higher probability 4) Go back to 2) Genes f(x i ) code Traditional Genetic Algorithm Genotype S S 2 n p!!! 1 S Code: φ x1 x 2 x n p Phenotype Variation Selection

17 biased population generation probabilistic selection Solution space encoded as finite-length string over a finite alphabet E.g. {0, 1} GA s exploit coding similarities Searches the code space, not the solution space Searches the space with many alternatives in parallel Avoids getting trapped in local optima Higher probability of finding better solutions Not random search Search towards regions with likely improvement Better solutions reproduce more often Does not work in very rugged, chaotic, uncorrelated landscapes Traditional Genetic Algorithm Genotype S S 2 n p!!! 1 S Code: φ x1 x 2 x n p Phenotype Variation Selection

selection reproduction 1) Reproduction: New population is generated a) Selection Select two parent chromosomes from a population according to their fitness Biased roulette wheel according to fitness

18 selection reproduction 1) Reproduction: New population is generated a) Selection Select two parent chromosomes from a population according to their fitness Biased roulette wheel according to fitness of solution Elite group Tournament f(x 3 ) f(x 2 ) f(x 1 ) code Traditional Genetic Algorithm Genotype S S 2 n p!!! 1 S Code: φ x1 x 2 x n p Phenotype Variation Selection

19 Variation: crossover reproduction 1) Reproduction: New population is generated a) Selection Select two parent chromosomes from a population according to their fitness b) Variation: Crossover With a crossover probability produce offspring pair by recombining parents Parents

20 Variation: mutation reproduction 1) Reproduction: New population is generated a) Selection Select two parent chromosomes from a population according to their fitness b) Variation: Crossover With a crossover probability produce offspring pair by recombining parents. c) Variation: Mutation With a mutation probability mutate (bit flip) new offspring at each bit position Parents

The workings 1) Generate Random population of bit-strings 2) Evaluate Fitness Function for each decoded solution 3) Reproduce next generation Selection by fitness Variation crossover and mutation

21 The workings 1) Generate Random population of bit-strings 2) Evaluate Fitness Function for each decoded solution 3) Reproduce next generation Selection by fitness Variation crossover and mutation Fill new population 4) Go back to 2) until stop criteria is met Desired fitness Specified number of generations Convergence Lack of variability in population and/or fitness Tends to a peak f(x 3 ) genetic algorithms f(x i ) f(x 2 ) f(x 1 ) Parents

22 artificial genotype/phenotype mapping Search algorithms based on the mechanics of Natural Selection Based on distinction between a machine and a description of a machine Solution alternatives for optimization problems Genotype DNA Inherited variation computational evolution transcription translation (code) development RNA amino acid chains Traditional Genetic Algorithm Genotype environmental ramifications phenotype organism S S n 2 p!!! 1 S Variation Code: φ code x x x 1 2 n p Phenotype Selection

readings Next lectures Class Book Nunes de Castro, Leandro [2006]. Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications. Chapman & Hall.

Chapter 2: The logical Mechanisms of Life Chapter 3: Formalizing and Modeling the World Chapter 4: Self-Organization and Emergent Complex Behavior Chapter 5: Reality is

23 readings Next lectures Class Book Nunes de Castro, Leandro [2006]. Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications. Chapman & Hall. Chapter 2, 7, 8 Appendix B Turing Machines, Computational complexity Chapter 3, all sections Sections 7.8, 8.3.2, Lecture notes Chapter 1: What is Life? Chapter 2: The logical Mechanisms of Life Chapter 3: Formalizing and Modeling the World Chapter 4: Self-Organization and Emergent Complex Behavior Chapter 5: Reality is Stranger than Fiction posted Optional materials Flake s [1998], The Computational Beauty of Life. MIT Press Chapter 20 Scientific American: Special Issue on the evolution of Evolution, January 2009.

biologically-inspired computing lecture 18

biologically-inspired computing lecture 18 Informatics -inspired lecture 18 Sections I485/H400 course outlook Assignments: 35% Students will complete 4/5 assignments based on algorithms presented in class Lab meets in I1 (West) 109 on Lab Wednesdays